ChatGPT and Claude do not show up in Google Analytics. That does not mean AI is not driving business. It means the old measurement playbook is broken and you need a new one. Here is what actually works.
When someone clicks a link from a Google search to your site, your analytics tool sees a clean referrer header. You know the visit came from google.com, you often know the search query, and you can attribute the conversion to a particular page. That entire chain breaks for AI traffic.
ChatGPT, Claude, Perplexity and Gemini either strip the referrer header entirely or pass one that gets bucketed as direct traffic. The user reads an answer, sees your brand mentioned, then either types your URL into their browser, searches for your name on Google, or clicks a link that arrives at your site with no breadcrumb back to the AI. The result is that AI-driven revenue is showing up in your analytics as "direct" or "organic" with the wrong query, and most marketing teams are quietly under-attributing it.
This is not a new problem. Dark social, the offline-to-online gap, podcast attribution, all of these have the same structure. The trick is to stop trying to force AI traffic into the old funnel chart and instead build a measurement stack designed for an ambient, citation-driven channel. If you want context on why this channel is now too big to ignore, see why GEO matters now.
Four signals together give you a defensible picture of GEO performance. None of them is perfect on its own. Combined, they tell a clear story.
Google Search Console is your friend here. Pull the impressions and clicks for your brand name and obvious variants over the last 12 months. AI citations push users to verify by Googling the brand. A sustained rise in brand-name searches without matching paid spend, PR coverage, or product launches is a reliable proxy for AI visibility. Track this monthly.
Direct traffic to deep pages, especially pages a user would not memorise, is highly suspicious in a good way. A spike in direct visits to a comparison page or a feature page, with no email or paid campaign running, almost always means AI is citing the URL. Look at direct traffic to non-homepage URLs as its own metric.
Pick 50 to 100 buyer-intent prompts that matter for your business. Run them through ChatGPT, Claude, Perplexity and Gemini on a weekly schedule. Record whether your brand was cited, in what position, and against which competitors. This is the closest thing GEO has to a keyword ranking report. Free and paid tools exist for this, or you can script it. We run this exact loop for every AI visibility check we do.
Add an optional "how did you first hear about us" field to your post-signup or post-purchase flow with "AI assistant (ChatGPT, Claude, etc.)" as one of the named options. The data is noisy, but the trend line over six months is more reliable than any clever attribution model. Even at low response rates, the directional signal is clear.
The most important new measurement discipline in GEO is prompt position tracking. The mechanics are simple: define a set of prompts you care about, run them through the major AI assistants on a recurring schedule, and log the answers in a database. Over time you get a structured time series of how each AI cites your brand, your competitors, and the questions you most want to win.
The questions you choose matter enormously. A SaaS company should track "best [category] for [use case]" prompts plus "alternatives to [largest competitor]". A local service business should track "best [service] near [city]" plus "[service] recommendation in [neighbourhood]". A consultancy should track "who is the best [specialism] consultant for [industry]" plus "top experts in [niche]". The goal is to choose 50 to 100 prompts that, if AI cited you in answer to every one of them, would meaningfully change your business. For more on how this fits into the broader picture, see how AI recommends businesses.
Once you have the four-signal stack running, you can start to model revenue contribution. The cleanest approach is a holdout-style test. Pick a target prompt set, invest in moving your citation rate on those prompts over a quarter, and watch what happens to branded search, direct traffic to relevant pages, and self-reported AI attribution. If all three move together in the right direction while paid spend stays flat, you have a defensible attribution story.
Do not chase a perfect attribution model. The point of GEO measurement is not pixel-accurate revenue allocation. It is to know whether the channel is working, which prompts are paying off, and where to invest next quarter. A messy but consistent measurement loop beats a perfect one you never run.
The biggest measurement mistake teams make is waiting until they can perfectly attribute GEO revenue before investing in it. By the time perfect attribution exists, your competitors will have a two-year head start. Start the four-signal loop now, accept the imperfection, and let the trend lines do the talking.
Treat GEO measurement as a parallel track to your existing SEO and paid reporting, not a replacement for either. The differences between the two disciplines are covered in GEO vs SEO, and the underlying mechanics in how GEO works. The dashboards live side by side. Over time, as AI assistants capture more of the discovery journey, your GEO dashboard will become the leading indicator for the others. Brand search lift, direct traffic to deep pages, prompt position, and AI-attributed signups are the new metrics that matter.
Run a free AI visibility check and see exactly which prompts are citing you today.
Run a Free AI Visibility Check